Visiting Distant Neighbors in Graph Convolutional Networks

In this study, we expand the graph convolutional network layers for deep learning on graphs to higher order in terms of neighboring nodes. As a result, when constructing representations for a node in a graph for downstream tasks, in addition to the features of the node and its immediate neighboring nodes, we also include more distant nodes in the aggregations with tunable importance parameters. In experimenting with a number of standard benchmark graph datasets, we demonstrate how this higher-order neighbor visiting pays off by outperforming the original model especially when we have a limited number of available labeled data points for the training of the model.